Вы находитесь на странице: 1из 32

Agreement in Arabic

The last chapter examined in detail the syntactic behaviour of the Finnish
possessive suffixes both in the possessive construction and a number of non-

finite verbal clause-types. Taken at face value, the data suggested that the nominal properties
of these suffixes (specifically whether they should be analysed as agreement morphemes or
pronominal affixes and their binding-theoretic status) varied according to the construction in
which they appeared, leading some scholars to propose, undesirable as it may at first seem,
that the language has two sets of morphosyntactically distinct morphemes which happen to be
phonetically identical. On the basis of correlations between restrictions on word order and the
properties of the suffixes in each of the constructions under discussion, this step was shown to
be unnecessary. An alternative was developed, according to which the -features of the affix
were always lexically valued, but their status with respect to LF-interpretability determined at
the point at which they are introduced into the derivation: taking affixes merged in a
-position to have interpretable -features and those merged in a non--position to have
uninterpretable -features in need of deletion was shown to account for the full range of data

in a more principled and elegant way than any of the three alternative analyses considered. By
drawing on ideas familiar from standard analyses of the passive, the ability of a verb to assign
an external -role was argued to be inhibited by the morphology associated with some nonfinite verb forms, but not others.

This chapter and the next will explore the more general validity of these claims by
considering languages with similar agreement patterns and examining the extent to which the
model proposed can account for these. The first such language is Modern Standard Arabic, in
which, as Fassi Fehri observes, bound nominative forms are homonymously ambiguous
between a pronoun and an inflection interpretation (Fassi Fehri, 1993:121). This situation is
reminiscent of the lexical split proposed by Toivonen to account for the different behaviour of
the Finnish possessive suffixes (cf. section 2.3.3 of the last chapter) and it is the aim of this
chapter to show that it is amenable to an analysis along similar lines. Section 1 introduces the
two basic word orders of Modern Standard Arabic finite clauses and presents the restrictions
on agreement associated with each. Section 2 considers the syntax of SVO sentences, showing
it to be directly comparable to that of the Finnish temporal adjunct and agent constructions,
and in so doing refines the model of -role assignment hitherto adopted. After object and
126

complementiser agreement have been shown (in section 3) to have essentially the same
properties as the Finnish participial construction, section 4 tackles the problem of partial
agreement in VSO clauses, a phenomenon for which there is no direct parallel in Finnish. In
spite of this, it is shown that such clauses can be accommodated in the model by adopting
Mohammads (1989, 1990) Expletive Hypothesis for VSO orders. Section 5 considers verbinitial sentences with pronominal subjects, which do display full agreement and argues that
these are not true arguments, but rather occupy an -position, from where they bind the
pronominal affix. Section 6 considers why these additional options are not be available in
Finnish and section 7 concludes the discussion.

PATTERNS OF AGREEMENT IN MODERN STANDARD ARABIC

Modern Standard Arabic allows two basic word orders in finite declarative clauses with
different degrees of subject agreement being possible in each. In SVO clauses, the verb
obligatorily agrees with an NP-subject in all -features (1), while in VSO clauses only gender
agreement is possible (2). Partial (i.e. gender only) agreement is not possible in SVO clauses
(3), nor is full agreement possible in VSO clauses (4)1.

(1)

n-nis?-u

daxal-na

maktib-a-hunna

the-women-NOM enter.PAST-3PL.F office.PL-ACC-their.F


The women have entered their offices

(2)

daxal-at

n-niss?-u

(Fassi Fehri, 1993:32)

maktib-a-hunna

enter.PAST-3SG.F the-women-NOM office.PL-ACC-their.F


The women have entered their offices

(3)

*n-nis?-u

daxal-at

(Fassi Fehri, 1993:32)

maktib-a-hunna

the-women-NOM enter.PAST-3SG.F office.PL-ACC-their.F


Intended: The women have entered their offices

(4)

*daxal-na

n-niss?-u

(Fassi Fehri, 1993:32)

maktib-a-hunna

enter.PAST-3PL.F the-women-NOM office.PL-ACC-their.F


Intended: The women have entered their offices
1

(Fassi Fehri, 1993:32)

In the interests of consistency, some of the Arabic examples are reproduced here in a form slightly different

from that in which they appear in the sources cited.

127

Arabic allows pronominal arguments of all kinds to remain unexpressed in the presence of a
corresponding agreement morpheme on the verb. Subject and object agreement differ in this
respect to the extent that, while an overt subject may co-occur with subject agreement, where
the verb carries object affixes, the argument they cross-reference is obligatorily null2.

(5)

(?an)

intaqad-tu-hu

(*?iyyhu)

(I.NOM) criticise-PAST.1SG.NOM-3SG.M.ACC3 (*he.ACC)


I criticised him

(6)

(?anta)

?at.ay-ta-n-hi

(*?iyyya) (*?iyyh)

(you.SG.M.NOM) give-PAST.2SG.M.NOM-1SG.ACC-3SG.F.ACC (*I.ACC)

(*she.ACC)

You gave her to me

(7)

(hunna)

i?-na

(they.F.NOM) come-PAST.3PL.F.NOM
They came

A pronominal object must be expressed as an affix on the verb rather than as a free form
wherever this does not violate another principle of the grammar. One such principle is the
person constraint, which requires object affixes to appear in ascending order of value of
person feature. However, the order of the affixes is also constrained by semantics, with the
first of two object affixes attached to a ditransitive verb obligatorily being interpreted as the
indirect object and the second as the direct object. The conflict between the requirements on a
verb with, for example, a second person direct object and third person indirect object is
resolved by expressing one of the arguments as a free form.

(8)

*?at.-hu-ka

l-?ustd-u

give-PAST.3SG.M.NOM-3SG.M.ACC-2SG.M.ACC the-teacher-NOM
Intended: The teacher gave you to him

(Fassi Fehri, 1993:104)

This is a slight oversimplification (see sections 4.3.2 and 5 below).

Anticipating the proposal in section 5.1 that agreement affixes have Case, subject and object agreement will be

glossed throughout this chapter as NOM and ACC respectively.

128

(9)

?at.-hu

l-?ustd-u

?iyyka

give-PAST.3SG.M.NOM-3SG.M.ACC the-teacher-NOM you.SG.M.ACC


The teacher gave you to him

(Fassi Fehri, 1993:105)

The object affixes also appear on certain complementisers such as ?inna, cross-referencing the
-features of the subject (10) and in this respect behave in the same way as a matrix predicate
exceptionally Case-marking the subject of the embedded clause (11). In both these cases too,
the affixes are in strictly complementary distribution with overt arguments.

(10)

qla

?inna-hu

?-a

say.PAST.1SG.NOM that-3SG.M.ACC come-PAST.3SG.M


I said that he came

(11)

?-a

hsib-tu-hu

think-PAST.1SG.NOM-3SG.M.ACC come-PAST.3SG.M
I thought he came

(12)

qla

(Fassi Fehri, 1993:98)

?inna(*-hu)

l-?ustd-a

?-a

say.PAST.1SG.NOM that-3SG.M.ACC the-teacher-ACC come-PAST.3SG.M


I said that the teacher came

(13)

hsib-tu(*-hu)

l-?ustd-a

?-a

think-PAST.1SG.NOM(*-3SG.M.ACC) the-teacher-ACC come-PAST.3SG.M


I thought he came

2 AN ANALYSIS OF SVO ORDER


2.1 The Structure of SVO Clauses
The restrictions on the co-occurrence of agreement and overt subjects in Arabic finite clauses
with SVO word order are exactly the same as those observed for first and second person
subjects in the Finnish temporal adjunct and agent construction in section 2.1 of the preceding
chapter. Verbs exhibiting full agreement may be preceded by pronominal subjects without
these necessarily needing to receive focal or contrastive stress of any kind.

129

(14)

(?an)

ra?ay-tu

zayd-an

(I.NOM) see-PAST.1SG.NOM Zayd-ACC


I saw Zayd

(15)

(?anti)

takallam-ti

(you.SG.F.NOM) speak-PAST.3SG.F.NOM
You (f.) spoke

It is only in the third person that the constructions in the two languages have different
properties in this respect. Firstly, as a comparison of (1) and (7) reveals, full NPs in Arabic,
unlike their counterparts in Finnish non-finite clauses, trigger the same agreement as third
person pronouns. Secondly, whereas in Finnish null third person subjects are anaphoric to the
extent that they are only grammatical when co-referential with an antecedent in a higher
clause, in Arabic they may occur in exactly the same range of contexts as null first and second
person subjects.

(16)

(huwa)

?-a

(he.NOM) came-3SG.M.NOM
He came

(Fassi Fehri, 1993:115)

Drawing from these observations the conclusion that Arabic differs from Finnish in that full
NPs are specified for -features4 and third person agreement licenses a null pronominal in the
same way as first and second person affixes do5, it is a simple matter to apply the model
developed for the Finnish temporal adjunct and agent construction directly to Arabic SVO
clauses. This amounts to saying that the topmost node of the verbal complex (presumably v in
a finite clause) does not assign a -role to its left with the consequence that, by the proposal in
3.4.2 of the last chapter, the -features of any agreement affix Px6 merged with vP will not be

Note that while the Finnish data suggest that having -features (or, more specifically, a person feature) is

sufficient for a category to be able to receive a -role, it does not support the view that this is necessary, if the
suggestion (made in section 2.1 of the last chapter) that full NPs are not valued for person is correct.
5

As yet no precise account of how optional arguments are licensed has been given. This is the subject of chapter

six.
6

In order to make the parallels clear, I shall continue to use the symbol Px to denote an agreement head.

130

interpretable and must therefore be deleted. At this stage in the derivation then, the structure
of the clause is as follows7.

(17)

PxP
qp
Px
vP
uval
qp
v

The derivation of the Finnish constructions then proceeded by introducing the head , which
word-order evidence from clauses of the same type with no possessor agreement had already
shown to trigger movement of the verbal complex out of VP. The subject was then merged in
the specifier of , checking the uninterpretable -features of Px and receiving genitive Case.
Assuming the analogue of this to be the nominative Case-assigner T, the structure of finite
Arabic SVO clauses is expected to be the following (cf. (118) of the chapter three).

(18)

TP
qp
SUBJECT
qp
[CASE: ] T-v-Px
PxP
ival
uval
qp
assignNOM
tPx
vP
qp
tv

2.2 The Locus of -Role Assignment in SVO Clauses


One claim of this analysis is therefore that preverbal subjects in finite SVO clauses originate
in SpecTP and as such seems to be at odds with the general, if not unanimous, consensus that
SpecvP is the position in which thematic subjects are introduced into the structure. While it is
true that Doron and Heycock (1999) have argued convincingly that a class of elements with
subject properties are first merged in SpecTP8, these differ from the kinds of subject under
discussion here in being associated with a -role assigned further down the structure and
hence rely crucially on SpecTP not being a -position. This will clearly not be the case for the
structure shown in (18), if the position of first merge is always the position in which an
element receives its -role. Furthermore, anticipating the discussion of VSO orders in section
7

This is essentially the same as that given for the temporal adjunct and agent constructions in (129) of the last

chapter.
8

These data will be considered in detail in section 4.3.2.

131

4 below, there seems to be ample evidence that v can indeed assign the subject -role in
Arabic finite clauses.

In Finnish, the pattern of agreement exhibited in a given context was dependent on the
construction in which it occurred and as such, whether or not the verb was able to assign a
subject -role could safely be assumed to be determined by a morphosyntactic property of the
affixes (va/nut-, de-, ma- etc.) making up the form in question. This line of reasoning is not
available for the Arabic data, as it is only the patterns of agreement, rather than the forms of
the verb themselves that differ according to word order. What the data seem to suggest is that
v may, but need not, assign a -role to the category immediately c-commanding it, that -role
remaining available for assignment to a category in SpecTP once v-to-T movement has
occurred. This is, of course the same conclusion as was reached on the basis of purely
theoretical considerations in section 2.2.4 of chapter two and the Arabic data may therefore be
taken to constitute independent empirical support for this position. Moreover, it also resolves
an anomaly in the analysis proposed in the last chapter, where it was proposed that the Caseassigner was also a -role assigner in the temporal adjunct and agent constructions but not
in the participial construction. To all intents and purposes, this amounted to proposing the
existence of two distinct heads, albeit with overlapping properties, but under the new
proposal, even where the -role is assigned from , it need not be assigned by . This head
can thus be taken to have the same properties in both constructions (with any further
restrictions on where -roles can be assigned in different types of clause still being taken to
originate in the morphology). An SVO clause is therefore simply the structure that emerges
when the verb waits until is has moved as far as is possible before assigning its subject -role.
This model will be assumed from this point onwards and the phrase assigned from a head
rather than assigned by a head used to identify the position in which an assigning head
discharges its -role.

However, while resolving this one issue, adopting this analysis raises the question of whether
the -role can also be assigned from Px, through which the -assigner clearly moves where it
is present. In the analysis proposed for the Finnish temporal adjunct and agent construction
and now adopted for finite SVO clauses in Arabic, it was assumed that Px first moves to /T
and has its -features checked there by an element merged in SpecP/SpecTP rather than
allowing the subject to be first merged in SpecPxP, but no support was offered for this
conclusion. Indeed, since Px in these constructions has uninterpretable -features, there is
132

ample motivation for merging the subject in SpecPxP and no a priori reason why this should
not be allowed. If it were, however, then sentences such as (19) in Finnish and (4) in Arabic,
repeated here as (20), should be grammatical, the verb raising in both cases past the subject in
SpecPxP to check tense and the object kattoa roof being targeted by the EPP-feature on in
the Finnish example, but it would be reasonable to contend that in the absence of
corroborative evidence for such an analysis from other constructions, the stipulative nature of
such a restriction reduces the explanatory power of the account as a whole.

(19)

*Katto-a korja-te-ssa-ni

minu-n Nelli

roof-PAR fix-INF2-INE-1SG.PX I-GEN

l-i

sormee-nsa

Nelli[NOM] hit-PAST[3SG] finger-3.PX

Intended: While I was fixing the roof, Nelli hit her finger

(20)

*daxal-na

n-niss?-u

maktib-a-hunna

enter.PAST-3PL.F the-women-NOM office.PL-ACC-their.F


Intended: The women have entered their offices

(Fassi Fehri, 1993:32)

As it happens, Arabic offers independent empirical support for the conclusion that the
subjects (of SVO sentences) do not originate in SpecPxP9, even if it does not necessarily offer
any further insight into the theoretical motivation behind the restriction. While the subjects in
all the examples considered so far have been nominative, this appears to be the option chosen
only when no other Case-assigner able to target the preverbal position is present in the
structure. If the clause is embedded under certain matrix verbs or introduced by certain
complementisers including ?inna (an option not restricted to embedded clauses), then the
subject must appear in the accusative.

(21)

haUsib-tu

baqarat-an/*-un

think-PAST.1SG.NOM cow-ACC/*-NOM
I thought that a cow has spoken

(22)

takallam-at
speak-PAST.3SG.F
(Fassi Fehri, 1993:48)

?inna baqarat-an/*-un takallam-at


that cow-ACC/*-NOM speak-PAST.3SG.F
A cow has spoken

(Fassi Fehri, 1993:48)

This position will be modified somewhat in section 5

133

If, as it seems reasonable to assume, a Case-assigner assigns its Case to the nearest Caseless
NP in its c-command domain, then this result is unexpected if the subject originates in a
position below T, such as SpecPxP (or SpecvP for that matter). In that case, the subject would
be expected always to receive nominative Case and consequently be unable to receive the
accusative Case assigned by any complementiser or matrix verb present after moving to the
preverbal position (23). (This concurs with the fact that postverbal subjects in VSO clauses
are invariably nominative (24), even when the clause is introduced by an accusative-assigning
head, although as will be seen in section 4.3, the situation is a little more complicated.)

(23)

(24)

CP/VP
qp
C/V
TP
assignACC
qp
SUBJECT
qp
[CASE:NOM]
T
PxP
assignNOM
qp
tSUBJECT
qla

?inna-h

?-at

l-bant-u/*-a

say.PAST.1SG.NOM that-3SG.F.ACC arrive-PAST.3SG.F.NOM the-girls-NOM/*-ACC


I said that the girls arrived

(Mohammad, 1999:143)

If, on the other hand, the subject is first merged above T, then it is not in its c-command
domain and can escape being assigned nominative Case in the presence of another Caseassigner. If no such head is available then nominative is assigned as a last resort, either by T
or by means of a default rule.

At this point the question arises as to why it is that the subjects of Finnish constructions
claimed here to have essentially the same structure as in (18) can have no other structural
Case than genitive. This is easily accounted for if we recall that the temporal adjunct, as its
name suggests, is only ever found in adjunct positions and are as such not within the
c-command domain of a structural Case assigner, with the result that the clausal head
always assigns genitive as a last resort. This explanation might, at a first glance, appear to be
falsified by the participial and agent constructions, which do appear to occupy a position
accessible to a Case-assigner.

134

Where an agent construction modifies a direct object, as is the case in the partial structure
(25), its subject is within the c-command domain of the verb and should therefore be able to
receive accusative or partitive Case. (26) shows that this prediction is incorrect: only the
genitive is grammatical.

(25)

(26)

VP
qp
V
DP
mru
assignACC/PAR
P
4
ei auton
hneei
[CASE: ]
PxP
osta-ma-nsa
Min m-i-n

hne-n/*-t/*-a

osta-ma-nsa

auto-n

I[NOM] sell-PAST-1SG I-GEN/*-ACC/*-PAR buy-INF3-3.PX car-ACC


I sold the car that he bought

Similarly, in the partial structure (27), the subject of the embedded participial clause is
immediately c-commanded by the matrix verb muista- and should therefore also be able to
receive the accusative or partitive Case canonically assigned to object positions. (28) shows
that it cannot.

(27)

(28)

VP
qp
V
P
muistaqp
assignACC/PAR minuqp
[CASE: ]

VP
soitta-neeEero

muista-a

minu-n/*-t/*-a

soitta-nee-n

haitaria-a

Eero[NOM] remember-3SG I-GEN/*-ACC/*-PAR play-NUT-DFT accordion-PAR


Eero remembers me playing the accordion

The explanation of the ungrammaticality of (25) is straighforward. Assuming that the verb
can assign structural Case to a maximum of one category, this will always be to the head of
the object DP auton. If this Case were assigned to the subject of the agent construction, there
135

would be no other way for the object to get Case and the structure would crash. The subject of
the agent construction, on the other hand, is accessible to the Case-assigner , with the
consequence that only where this option is chosen do grammatical sentences result.

Turning now to the participial construction, recall from section 3.4.1 of the last chapter that
the participial construction was argued to differ from the temporal adjunct and agent
constructions in that an overt subject was merged in SpecVP (where V is the highest head of
the verbal complex). When merged as the object of a verb assigning accusative or partitive
Case, it will have the structure given in (29), which is analogous to (23). Unlike that of the
temporal adjunct and agent constructions, the subject in this structure is always in the
c-command domain of from which it will receive genitive Case and consequently be unable
to receive Case from V.

(29)

VP
qp
V
P
assignPAR/ACC
qp
SUBJECT
qp
[CASE:GEN]

VP
assignGEN
qp
tSUBJECT

It could of course be argued that while V is clearly able to assign its -role to the category in
SpecVP, there should be nothing to prevent it waiting until it has moved to to do so, thereby
making the subject available for exceptional Case-marking by the matrix verb. In the
discussion of this model of -role assignment in section 2.2.4 of chapter two, it was argued
that the choice of whether to merge the subject or the Case-assigner first was arbitrary, since
both satisfied a requirement of the topmost head of the verbal complex (the subject enabling a
-role to be discharged, the Case-assigner T fulfilling the verbs need for tense), with the
consequence that both orders could be derived without movement. In the case of the
participial construction, however, the verbal complex is not dependent on in the same way,
as the grammaticality of participial clauses where this head is absent altogether shows. The
only way the derivation can proceed following the merger of the topmost head of the verbal
complex V is therefore for the subject to be introduced.

136

By hypothesis, then, if the Finnish temporal adjunct could be embedded under verbs or
complementisers assigning a structural Case, then their subjects would have that Case in those
contexts rather than the canonical genitive. However, the structures of such non-finite clauses
(with overt subjects) as do occur in these contexts are such that either there is another
category which must receive the structural Case in question from the verb in order for the
structure to be grammatical, or that the preverbal subject is already marked as genitive by the
time it arrives in SpecP. Thus different factors conspire to prevent subjects of Finnish
non-finite clauses from displaying the same range of Cases as those of Arabic SVO clauses.

OBJECT AND COMPLEMENTISER AGREEMENT

The restrictions on the co-occurrence of agreement and overt arguments in the case of object
and complementiser agreement, on the other hand, are the same as those applying to the
Finnish participial construction and an account akin to the one offered in section 3.4 of the
last chapter can be successfully applied here. There it was argued that the highest head of the
verbal complex was a -role assigner with the consequence that the -features of an overt
subject or agreement head Px merged with its projection would be interpretable. By the same
token, if the subject -role in an SVO clause is assigned by T, any Px head merged
immediately above it will have interpretable -features and function as a pronoun.

(30)

CP
wo
C
PxP
wo
Px
TP
[ival]
wo
T
assign

Where SpecTP is filled, the element in that position will receive the -role from T, such that
any Px subsequently merged would have uninterpretable -features, given the conclusions
reached in section 2.4.2 of chapter three. Being valued, these features would be unable to
probe the subject in SpecTP, the only element able to check them, and remain undeleted,
resulting in ungrammaticality (31). Only where no such head is merged can the derivation
converge.

137

(31)

CP
wo
C
PxP
assignACC
wo
Px
TP
* [uval]
wo
SUBJECT
wo
[CASE: ]
T
[ival] assign

Similar reasoning applies to where these affixes function as thematic objects, the only
difference being that it is V rather than T that assigns the -role and v rather than C that
assigns Case to an overt object.

4 THE PROBLEM OF PARTIAL AGREEMENT IN VSO ORDERS


While the patterns of subject agreement in SVO clauses and of object and complementiser
agreement in general can be accommodated in the model developed for Finnish non-finite
clauses, the pattern of subject agreement in Arabic VSO clauses is different from that found in
either kind of clause discussed in chapter three. As long as gender is ignored, the
co-occurrence of agreement and overt subjects in Arabic VSO clauses is subject to similar
restrictions as the Finnish participial construction. Extending this analysis developed in
section 2.4 of the previous chapter for such clauses to the Arabic data predicts correctly that
an agreement affix merged directly with the verbal complex can receive a -role from v and
function as an argument and also accounts for the lack of number agreement with the
postverbal subject in examples (1) and (2) above. However, these same examples show that it
runs into difficulty as soon as gender is taken into account, for it is clear that the form of the
verbal affix varies according to whether the subject is masculine or feminine. In this respect,
such default agreement in Arabic VSO clauses is crucially different from the default
agreement found in the Finnish participial construction. For while the ending n, found in the
latter in the presence of an overt subject, is distinct from any of the possessive suffixes and
can therefore reasonably be assumed to have no -features, the situation in Arabic is not as
clear cut, since verbs in VSO constructions carry the same agreement as that triggered by
third person singular preverbal subjects. It could, of course, be claimed that the difference
between the two languages in this respect is purely morphological in nature: in Arabic (but
not Finnish), the verbal affixes encoding lack of -features just happen to have the same
phonetic form as those specified as [3SG]. However, since it was shown in the previous
chapter that even the Finnish possessive suffixes, the confusing syntactic behaviour of which
138

had led commentators to posit a functional ambiguity of precisely this type, were amenable to
an elegant uniform analysis, the hypothesis should at least be explored that Arabic third
person singular agreement morphemes have the same lexical specification in both the contexts
in which they occur.
One consequence of adopting this hypothesis is that Arabic VSO constructions instantiate a
clause type argued to be impossible in the last chapter, namely one in which a Px head is
allowed to co-occur with a subject that is indisputably merged low (in SpecvP in this case).
Based on the assumption that only unvalued features are able to probe, it was shown there that
merging a Px head above an overt subject would always lead to ungrammaticality, because
the valued -features of the Px head could not probe the subject and would thus not be
deleted, while any category merged above Px to this end would be unable to receive a -role.
If this analysis is to hold for Arabic therefore, it must be shown either that the -features of
Px do not need checking in VSO clauses, or that the postverbal subject is in a position to
check them after all, or that another element is available to check them (and that this element
does not need a -role).

4.1 Third Person Singular Features Do Not Need Checking in VSO Clauses
Exploring the first of these possible explanations, the simplest reason for a feature not to need
deleting is that it is simply not present in the first place. Since the clauses under consideration
only ever have third person singular features, this amounts to saying that third person is not a
possible -feature value in Arabic, but is in fact the reflex of the absence of a person feature,
and singular number similarly is the absence of a number feature. However, while proposals
along these lines have been made for a number of other languages, there is good reason to
suppose that they are not compatible with the hypothesis being explored here. Firstly, if a
given Px head is to have the same features lexically, regardless of whether these are
interpretable or not, then this analysis makes incorrect predictions about the interpretation of
sentences with null third person subjects. For while the case can be made that the -values of
the Px head for person and number in VSO structures are underspecified, this reasoning
cannot be extended to gender, which does co-vary with that of a postverbal subject. The
-specification for the Px head must therefore include at least a valued gender feature, even in
VSO structures. Now, if v can assign a -role to an overt subject, it should also be able to
assign it to a Px head in the absence of such a subject, just as the topmost head of the verbal

139

complex was argued to assign a -role to either an overt subject or a possessive suffix in the
Finnish participial construction (cf. section 2.4.3 of chapter three).

(32)

PxP
qp
Px
vP
[ival]
qp
v
assign

By virtue of carrying this -role, Px functions as a bound pronoun and its -features are
interpretable. In the case of a third person singular affix therefore (if these are unspecified for
person and number as proposed above), the -features of Px will consist of a single gender
feature, [i:M] or [i:F]. As such, their interpretation should be broader than that of the lexical
pronouns huwa he and hiya she (which presumably have the feature matrices [i:3SG.M]
and [i:3SG.F] respectively), also including first and second person and dual and plural
subjects10. However, the subjects of (33) and (34) must be interpreted in the same way as third
person singular pronouns: the more general reading is not possible, suggesting that the
-features of the affixes in question are fully specified.

(33)

?-a
come-PAST.3SG.M.NOM
He came
*Someone/something masculine came

(34)

?-at
come-PAST.3SG.F.NOM
She came
*Someone/something feminine came

10

The notion that third person is in fact the absence of person could perhaps be upheld by arguing that, in the

absence of a clear indication of a discourse participant, a default non-participant value is assigned to the
subject. It is, however, less than obvious that third person is a discourse default. The subject of the English
sentence Went to the shops yesterday, for example, is naturally construed as first person, not third, in the absence
of an overt subject.

140

It could be argued that the absence of a person and/or number feature prevents the affix from
functioning as an argument and that the -features of a third person singular Px are therefore
always uninterpretable, regardless of whether they are introduced into the structure
immediately above a -assigner or not. If this is the case, then the status of Px is the same as
in SVO clauses, in which (according to the argument in section 2 above) the subject -role is
assigned by T rather than v and the -features of Px are always uninterpretable. If third person
singular morphemes are only specified for gender, then they should differ from the other
affixes in being compatible with preverbal subjects of all person-number combinations, not
only third person singular ones. Whereas the derivation of (35) would be predicted to crash
because the uninterpretable -features [1SG] of the morpheme tu would remain undeleted,
the absence of such features in (36) should remove the need for a subject with -values
matching the suffix a. This analysis thus predicts, incorrectly, that full agreement with a
preverbal subject, although possible, should not be obligatory.

(35)

*?anta

intaqad-tu

zayd-an

you.SG.M criticise-PAST.1SG.NOM Zayd-ACC


Intended: You criticised Zayd

(36)

*?anta

intaqad-a

zayd-an

you.SG.M criticise-PAST.3SG.M.NOM Zayd-ACC


Intended: You criticised Zayd

Furthermore, irrespective of whether the Px head carries person and number features in need
of deletion, the fact remains that it has a gender feature, which in VSO clauses will be
uninterpretable under the assumptions adopted here. Since there is no reason to think that
Arabic has a lexical nominal category specified only for gender, it would seem reasonable to
suppose that, given the strong evidence against an underspecification analysis of third person
agreement, the same element as checks the gender feature could also check person and
number.

4.2 Postverbal Subject Checks Features


If the conclusion is correct that third person affixes are fully specified, then the second
possible explanation for the unexpected grammaticality of VSO clauses containing a Px head
mentioned above (that the subject is, after all, able to check an uninterpretable gender feature)
141

is also unlikely to be workable. That is not to say, of course, that there is no position available
from which it could plausibly check the gender feature: as was observed in the discussion of
-role assignment in sentences with SVO word order above (section 2.2), SpecPxP is just
such a position, but as the discussion in that section also showed, there is no means of
preventing it from also checking the person and number features and receiving nominative
Case in that position, predicting incorrectly that full agreement in VSO clauses should also be
derivable (by subsequent movement of the verbal complex to T).

4.3 Another Element Checks Features


The only remaining possibility then, is that an element other than the subject checks the
verbal -features in VSO clauses. This element must be phonetically null, have third person
singular features and, since it is not in a position to receive a -role (the postverbal subject
having received that assigned by v) must still be a legitimate LF-object without one. This
conclusion is not without its problems, however, for the element in question is, to all intents
and purposes a null expletive, the existence of which is dubious on conceptual grounds, since
it contributes neither to the PF- nor the LF-interface11. The hypothesis is all the more suspect
for the fact that neither of the overt third person pronouns with the same -feature values as
this expletive would have to have can occupy the preverbal position in a VSO clause.

(37)

(*huwa)

?-a

tbb-u

l-malik-i

(*he.NOM) come-PAST.3SG.M physician-NOM the-king-GEN


The kings physician came

(38)

(*hiya)

?-at

(Mohammad, 1999:117)

tbb-at-u

l-malikat-i

(*she.NOM) come-PAST.3SG.F physician-F-NOM the-queen-GEN


The queens physician came

11

(Mohammad, 1999:117)

The status of the -features of expletives presents an interesting conundrum for the proposal advanced in the

last chapter that the interpretability of such features depends on their receiving a -role. Since an expletive does
not, by definition, carry a -role, its -features should be uninterpretable and need deleting. This is clearly not
compatible with any model which allows expletives to trigger agreement. For the purposes of the present work,
therefore, it must simply be accepted that expletives are anomalous to the extent that their -features do not need
deleting, despite not being associated with a -role.

142

4.3.1

The Expletive Hypothesis

Nevertheless, the proposal that it is an expletive element that controls agreement in verbinitial structures has had currency among Arabic linguists since it was first advanced by
Mohammad (1989, 1990). In a critique of this position, which he calls the Expletive
Hypothesis, Fassi Fehri (1993) concedes that, despite sentences such as (37) and (38) not
being grammatical, there are constructions in which third person pronouns function as
expletives, but that in these cases, the third person plural pronouns hum (39) and hunna (40)
are also possible. If silent expletives are licit in clauses of the type exemplified by (37) and
(38), then there can be no principled reason for disallowing those with plural number, again
predicting, counter to fact, that plural verbs should be possible in VSO clauses.

(39)

hum

l-und-u

they.M.NOM the-soldiers-NOM
It is the soldiers. Thats soldiers

(40)

hunna

(Fassi Fehri, 1993:40)

n-nis?-u

they.F.NOM the-women-NOM
It is the women. Thats women

(Fassi Fehri, 1993:40)

Mohammad (1999:Ch.4) responds to a number of objections to the Expletive Hypothesis,


accounting for the apparent counterexamples just mentioned by means of the Binding Theory
and the Case Filter. Observing that the plural forms are only grammatical in non-verbal
sentences, Mohammad suggests that hum in (39) and hunna in (40) are licensed in the same
way as pronouns in equative sentences with thematic subjects such as (41) and (42). In these
sentences (for reasons which Mohammad does not identify) co-indexation of the pronoun and
R-expression is possible, despite constituting a Principle C violation.

(41)

huwai

taUbb-u

l-malik-i]i

he.NOMi [physician-NOM the-king-GEN]i


He is the kings physician

(42)

hiyai

[tbb-at-u

(Mohammad, 1999:116)

l-malikat-i]i

she.NOMi [physician-F-NOM the-queen-GEN]i


She is the queens physician

(Mohammad, 1999:117)
143

Mohammads rather vague claim that whatever is responsible for licensing [(41) and (42)]
licenses [(39) and (40)] (Mohammad, 1999:117) seems to imply that the pronouns in (39)
and (40) are also licensed by virtue of being co-indexed with their associate and are hence in
fact thematic subjects, to a degree at least. Nominal categories in verbal clauses, by contrast,
are subject to Principle C, with the effect that co-indexation of the preverbal pronoun and
postverbal subject in (37) and (38) would result in ungrammaticality. If third person plural
pronouns are only licensed in contexts where they can be co-indexed with a plural
R-expression, then this explains why they may not duplicate postverbal subjects in these kinds
of clauses, but the ungrammaticality of overt singular pronouns functioning as expletives in
the same contexts is still unexpected, particularly if a null subject with identical features is
supposed to be licit.
Mohammad proposes that the difference in the grammaticality of overt expletives in equative
and verbal sentences is due to the fact that in equative sentences two nominative Cases are
available (for reasons that Mohammad does not discuss), one of which is assigned to the
R-expression, the other to the expletive. In the verbal sentences (37) and (38), by contrast,
only one nominative Case is available, which, having already been assigned to the postverbal
subject, cannot also license an overt expletive. If this is the correct conclusion, then it is to be
expected that an overt expletive will be licensed if another Case-assigner is available in the
structure. Recall from section 2.2 above that certain complementisers and verbs assign
accusative Case to the thematic subject in the SpecTP position of an SVO clause they take as
their complement and from section 1 that where an accusative argument is pronominal, it
appears as an affix attached to its Case-assigner (in this case the complementiser or matrix
verb). If the possibility of an overt expletive in that same position is dependent on whether or
not it can receive Case, then overt object suffixes, functioning as expletives, should be
possible in embedded clauses introduced by such heads. The following example shows that
this prediction is indeed borne out.

(43)

qla

?inna*(-h)

?-at

l-bant-u

say.PAST.1SG.NOM that-3SG.F.ACC arrived-PAST.3SG.F.NOM the-girls-NOM


I said that the girls arrived

(Mohammad, 1999:143)

Adopting the Expletive Hypothesis therefore provides an element other than the postverbal
subject, able to check the uninterpretable -features of Px in Arabic VSO clauses. What it
144

does not predict, however, is that the expletive necessarily has the same feature value for
gender as the postverbal subject. Since the verb is not dependent on the subject to delete its
uninterpretable -features, it should be possible for a feminine verb (with a feminine
expletive) to cooccur with a masculine subject and vice versa. As it happens, the second of
these possibilities has been noted by Arab grammarians where the verb is separated from the
thematic subject by another category (44-45) and Mohammad even gives two examples where
nothing intervenes (46-47).

(44)

wa-kna

l-il-yahd-i

fi bild-i

l-arab-i

and-was.3SG.M for-the-Jews-GEN in countries-GEN the-Arabs-GEN


liyt-un

karat-un

settlements(F)-NOM many-NOM
The Jews in the Arab countries had many emigrant settlements
(Mohammad (1999:119), quoting Cantarino (1974:84-85))

(45)

qad kna

yaskunu

l-irq-a

?umam-un

muxtalifat-un

qad was.3SG.M settle.3SG.M the-Iraq-ACC peoples(F)-NOM different-NOM


Different peoples had settled in Iraq
(Mohammad (1999:119), quoting Cantarino (1974:84-85))

(46)

lawl-hu

la-dU-a

?asm?-u

kar-in

mina

if.not.for-3SG.M.ACC la-lost.3SG.M.NOM names(F)-NOM many-GEN of


l-kutub-i

n-nafsat-i

the-books-GEN the-precious-GEN
If it had not been for him, the titles of many precious books would have been lost
(Mohammad, 1999:119)

(47)

naSar-a

h?ul?i l-awr

spread-3SG.M.NOM these

naw-an mina -aqfat-i

the-slave.girls type-ACC of

These slave girls spread a type of culture

the-culture-GEN
(Mohammad, 1999:120)

The precise mechanisms behind gender agreement with a postverbal subject are clearly more
complicated than the Expletive Hypothesis can account for on its own and the account offered
here makes no claim to be exhaustive in this regard.
145

4.3.2

Multiple Subject Constructions

There is however one kind of construction which appears to falsify the Expletive Hypothesis,
namely what Doron and Heycock (1999) call the multiple subject construction12, in which the
preverbal position is occupied by a non-subject argument.
(48)

hind-un

yuqbilu-ha

T-Tullb-u

Hind-NOM meet.3SG.M-3SG.F.ACC the-students(M)-NOM


The students are meeting Hind

(Doron and Heycock, 1999:71)

Doron and Heycock put forward a range of evidence that the preverbal nominal hind-un (to
which they refer as the broad subject, thus distinguishing it from the narrow subject
T-Tullbu) is first merged in SpecTP. Of particular interest for the discussion in hand is the
fact that broad subjects exhibit the same variation in Case-marking as do preverbal subjects,
argued in sections 1 and 2 also to originate in that position. In a sentence such as (48), in
which there is no head external to the clause that could assign Case to the broad subject, it is
nominative. When the clause is embedded under a matrix verb or one of the complementisers
that assign accusative, on the other hand, it must exhibit that Case.

(49)

dhanan-tu

hind-an

yuqbilu-ha

T-Tullb-u

thought-1SG Hind-ACC meet.3SG.M-3SG.F.ACC the-students(M)-NOM


I believed Hind to have been met by the students

(Doron and Heycock, 1999:73)

However, unlike preverbal narrow subjects, which obligatorily trigger full subject agreement
on the verb, a broad subject, whether nominative or accusative, triggers no agreement at all.
As (50) and (51) show, it cannot even control the gender feature, which must rather track that
of the postverbal narrow subject13, apparently calling into question the validity of the
conclusion that it is the preverbal expletive that controls agreement in straight VSO clauses.
12

As Doron and Heycock note, this construction is usually considered by Arabic linguists to be a case of left

dislocation. See their paper and references cited there for arguments for and against this position.
13

Note that all the -features of the broad subject are cross-referenced by the object agreement affix ha in this

example, contrary to what was said in section 1 about object agreement and overt arguments being in strictly
complimentary distribution. The reasons for this apparent exception will be discussed in the next section. What
is important for the discussion in hand is that the broad subject, despite occupying the same position as that from
which a narrow subject deletes the unvalued -features of the Px head immediately below T, cannot control the
subject agreement.

146

(50)

*hind-un

yuqbila-ha

T-Tullb-u

Hind(F)-NOM meet.3SG.F-3SG.F.ACC the-students(M)-NOM


Intended: The students are meeting Hind

(51)

*dhanan-tu zayd-an

yuqbilu-hu

n-nis?-u

thought-1SG Zayd(M)-ACC meet.3SG.M-3SG.M.ACC the-women-NOM


Intended: I believed Zayd to have been met by the women

Doron and Heycock do, however, provide one piece of evidence that suggests that the
Expletive Hypothesis may still be tenable, in spite of these apparent problems. They take the
fact that broad subjects are not restricted to VSO-type clauses as an indication that Arabic
allows multiple specifiers, noting that where both broad and narrow subjects precede the verb,
the narrow subject must follow the broad subject.

(52)

hind-un

?aT-Tullb-u

Hind-NOM the-students(M)-NOM

yuqbilu-una-ha
meet.3.M-PL-3SG.F.ACC

The students are meeting Hind

(53)

*?aT-Tullb-u

hind-un

(Doron and Heycock, 1999:83)

yuqbilu-una-ha

*the-students(M)-NOM Hind-NOM meet.3.M-PL-3SG.F.ACC


Intended: The students are meeting Hind

(Doron and Heycock, 1999:83)

The precise implementation of this restriction (which can be replicated in the framework of
the theory being developed here, even if the two are not directly compatible) is of secondary
importance to the fact that these examples demonstrate irrefutably that two preverbal
positions are available. If this is the case, then there is no reason why the second, that
occupied by the narrow subject controlling the agreement in (52), should not be occupied by
an expletive in (48). Unlike the VSO sentences considered so far, however, there is no way of
forcing the expletive in such a context to be phonetically realised, since the patterns of Casemarking in embedded multiple subject constructions show that only the broad subject in the
higher specifier of T appears in the accusative (54). The narrow subject obligatorily has
nominative Case as it would in a matrix VSO clause, predicting correctly that it should be
impossible to spell out an expletive merged in the same position (55) either as an agreement
affix or as a free pronoun.
147

(54)

?aT-Tullb-u/*-a

dhanan-tu

hind-an

Hind-ACC

the-students(M)-NOM/*-ACC

yuqbilu-una-ha

thought-1SG

meet.3.M-PL-3SG.F.ACC

I thought that the students were meeting Hind

(55)

dhanan-tu(*-hu)

hind-an

yuqbilu-ha

T-Tullb-u

thought-1SG(*-3SG.M.ACC) Hind-ACC meet.3SG.M-3SG.F.ACC the-students(M)-NOM


I believed Hind to have been met by the students

The fact that broad and narrow subjects may co-occur preverbally is therefore at least
consonant with the possibility that it is a null expletive located in the lowest specifier position
of T that checks the uninterpretable -features of Px in sentences such as (48) and (49), even
if there is no way of proving conclusively that it is present.

5 AGREEMENT WITH PRONOUNS


The sentences used so far as examples of partial agreement have all had full NP subjects and
have thus really shown only that number agreement with a postnominal subject fails.
Consideration of overt pronominal subjects reveals a different pattern. In such cases, full
agreement is not only possible, it is obligatory.

(56)

i?-tu

?an

come-PAST.1SG.NOM I.NOM
I came

(57)

*?-a

(Mohammad, 1999:121)

?an

come-PAST.3SG.NOM I.NOM
Intended: I came

(Mohammad, 1999:121)

Furthermore, doubling the affix in this way is not restricted to subject agreement. Object
affixes, whether they cross-reference a direct object (58), prepositional object (59) or a
possessor (60) can also be doubled by a pronoun in all the contexts in which they occur.

(58)

?-antaqid-u-ka

?anta

1SG.NOM-criticise-IND-2SG.M.ACC you.SG.M.NOM
I criticise you

(Fassi Fehri, 1993:114)


148

(59)

marar-tu

bi-hi

huwa

bi-?ax-hi

passed.by-1SG.NOM with-3SG.M.GEN he.NOM not with-brother-3SG.M.GEN


I passed by him, not by his brother

(60)

?-as?al-u

an

xabar-i-ka

(Fassi Fehri, 1993:114)

?anta

an

xabar-

1SG.NOM-inquire-IND about news-GEN-2SG.M.ACC you.M.NOM not about news-1SG.GEN


I am inquiring about your news, not about mine

(Fassi Fehri, 1993:114)

In this respect the data diverge from the pattern attested in the Finnish participial construction,
in which overt pronominal subjects do not trigger any agreement in -features on the
participle, and as such cast doubt on whether the schema developed for that type of clause in
the previous chapter (61) is the correct analysis for subject agreement in Arabic VSO clauses
and object agreement generally.

(61)

qp
CASE ASSIGNER
PxP
qp
Px
qp
[ival] -ASSIGNER

There are, however, important differences between pronouns in these positions and other
kinds of argument in Arabic (including preverbal pronominal subjects with the same phonetic
form). Fassi Fehri (1993:Ch.31.6), noting that they obligatorily carry focal or contrastive
stress of some kind (as (56) to (60) indicate) follows traditional Arabic grammar in treating
such pronouns as a class separate from those that function as non-focussed arguments. While
this move enables him to maintain the generalisation that postverbal subjects may trigger only
gender agreement, it comes at the expense of introducing a further functional ambiguity into
the grammar, since the phonetic exponents of the new class of strong pronouns that it creates
are identical to those of the class of weak pronouns, which may occur, unstressed, in
preverbal positions. The following sections will show that, like the functional ambiguity
purported to obtain in the case of Finnish possessor agreement, here too the differences
between the strong and weak pronouns need not be lexically stipulated, but can rather be
predicted from interactions between the grammatical features of lexical items allowed or
required by the structural configurations into which they are introduced.
149

5.1 The Base Position of Emphatic Subjects


As it stands, there is no position in the structure in (61) which could be occupied by a
pronominal argument duplicating the -features of the head Px. On the one hand, a subject
merged in SpecPxP would, under the theory of agreement developed in the last two chapters,
delete the -features of Px, which would consequently have to be uninterpretable. This
situation could only obtain if v did not assign a -role to its left in such constructions. A
pronominal subject could then be merged in SpecPxP, checking the -features of Px as
required, before the verbal complex moved to T to give the VSO word order. This would give
the correct results for pronominal subjects, but as it stands, does not explain why full NP
subjects cannot also be merged in this position thereby triggering full agreement in VSO
clauses. In the absence of an independently motivated condition limiting the class of elements
that can occupy SpecPxP to pronouns, then, this hypothesis will thus fall at the same hurdle as
the idea explored in section 2.2, that all preverbal subjects originate in this position.

If, on the other hand, such a subject were merged in the same place as a postnominal NP
subject (as specifier of the -assigner), then it would receive the -role shown as being
assigned to Px in (61), the -features of which would then be uninterpretable and need to be
deleted. This poses essentially the same problem as that which the Expletive Hypothesis was
invoked to solve in the previous section: the valued -features of Px are not able to probe an
argument lower down the structure, but this appears to be the only category present able to
check them. However, while proposing the presence of null third person singular expletives
with gender features to match those of Px was relatively unproblematic in the case of subject
agreement and could be motivated by reference to their overt counterparts in comparable
constructions, the fact that there is no restriction on the -feature values of the Px head in this
range of constructions makes a similar argument for these structures unworkable. A proposal
along these lines would require there to be first and second person expletives, a linguistic
category not attested overtly in any dialect of Arabic (or indeed of any known language).
Even if a convincing case could be made for the existence such a category, it would remain to
be explained why it should never be able to have different -features from the postverbal
subject, as was shown to be possible in at least some cases with gender agreement (even if the
rules underlying this laxness were not fully understood). Furthermore, it would require there
to be a set of object expletives, which the same tests as revealed the presence of subject
expletives in SpecTP can show not to be present in the specifier position of the accusative
Case-assigner v. Since a vP is always within the c-command domain of a T head, an expletive
150

in SpecvP14 should always be lexicalised (presumably as a subject agreement affix, given that
expletives embedded under an accusative Case-assigner were realised as object agreement
affixes). This expletive will only be able to fulfil its purpose of deleting the uninterpretable
features of the object agreement if it has the same -values, predicting incorrectly that an
overt pronominal object should in fact trigger two lots of agreement with the object, one
expletive subject affix and one object affix.
(62)

*zayd-un

?anta

antaqid-u-ta-ka

Zayd-NOM criticise-IND-3SG.M.NOM-2SG.M.NOM(EXPL)-2SG.M.ACC you.SG.M.NOM


Intended: Zayd criticises you
There is also evidence from Case-marking that these pronouns do not originate in the same
position as full NP arguments. For while a full NP argument must receive accusative Case
from a verb (63) or accusative-assigning complementiser (64) to its immediate left, a doubling
pronoun can only ever appear in the nominative (65-66). An obvious explanation for this is
that full NP arguments originate in a position where they can receive Case, whereas doubling
pronouns do not, the absence of an appropriate Case-assigner suggesting that their nominative
form is the result of a default rule.
(63)

?-antaqid-u

Hind-an/*-un

1SG.NOM-criticise-IND Hind-ACC/*-NOM
I criticise Hind
(64)

qla

?inna l-?ustd-a/*-u

?-a

say.PAST.1SG.NOM that the-teacher-ACC/*-NOM come-PAST.3SG.M


I said that the teacher came
(65)

?-antaqid-u-ka

?anta/*?iyyka

1SG.NOM-criticise-IND-2SG.M.ACC you.SG.M.NOM/*you.SG.M.ACC
I criticise you
(66)

qla

?inna-h

hiya/*?iyyh

?-at

say.PAST.1SG.NOM that-3SG.F.ACC she.NOM/*she.ACC come-PAST.3SG.F


I said that she came
14

This would only be possible in SVO clauses, in which the -role associated with v has been argued to be

assigned from T rather than from v.

151

By the same token, there is good reason to suppose that the verbal affixes are Case-marked,
since their realisation as subject or object affixes varies according to their structural position
in the same way as the Case of NP-arguments does. For just as the object NP in (63) and
embedded subject NP in (64) must be accusative, so also must a pronominal argument in the
same contexts be realised as an object affix.

(67)

?-antaqid-u-ka/*-ta
1SG.NOM-criticise-IND-2SG.M.ACC/*-2SG.M.SUBJ
I criticise you

(68)

qla

?inna-h/*-at

?-at

said.1SG.NOM that-3SG.F.ACC/*-3SG.F.SUBJ come-PAST.3SG.F


I said that she came

Comparing the structure of the relevant parts of these sentences with (63) and (64), it is clear
that the relationship that the Px heads realised as object agreement bear to the Case-assigners
v and C (69) is similar both to that of the full NP arguments (70) and to the relationship a Px
head realised as subject agreement bears to the Case-assigner T (71). This being the case, it is
a small step to posit that Px heads have an unvalued Case-feature and are realised as subject
agreement when nominative and object agreement when accusative.

(69)

CP/vP
qp
C/v
PxP
assignACC
qp
Px
TP/VP
[i:val]
qp
[CASE: ]
T/V
assign

(70)

CP/vP
qp
C/v
TP/VP
assignACC
qp
NP-ARGUMENT
qp
[CASE: ]
T/V
assign
152

(71)

TP
qp
T
PxP
assignNOM
qp
Px
vP
[i:val]
qp
[CASE: ]
v
assign

5.2 The Thematic Status of Postverbal Pronouns


Combining these two insights, a strong case can now be made that emphatic pronominal
arguments do indeed originate in SpecPxP, in spite of the arguments presented against that
position at the start of this section. The fact that the head of the complement of T itself needs
Case prevents the element in the specifier position of that head from being Case-marked15,
while the spec-head relation between the pronoun and the affix guarantees that their -feature
values will be the same. The question now arises whether this congruence of -features arises
in the same way as it does in SVO clauses (where a preverbal subject deletes uninterpretable
-features in T) or via a different mechanism not hitherto discussed and the answer to this
question has implications for the issue, examined in 2.2 above, of where -roles may and may
not be assigned. For if agreement with an emphatic pronoun is the result of it deleting the
uninterpretable -features of a Px head, it cannot be the case that the -role is assigned low in
these structures, as this would lead, in the absence of an element occupying the specifier
position of the -assigner, to the -features of Px being interpretable. The only alternative
explanation that is both compatible with this mechanism of agreement and also yields the
correct word order does therefore indeed seem to be that the -role is, just in this restricted set
of cases, assigned from Px to its specifier.

There are, however, good reasons to believe that this is not the correct analysis. Firstly, while
the -assigning heads in subject and object agreement configurations (v and V) do invariably
move through any Px head immediately dominating them, this is not the case with
complementiser agreement, where the -assigning head manifestly does not move any higher
than T and hence should not be able to assign its -role to any position above SpecTP.
Secondly, even in the case of subject and object agreement, arguments from the interaction of
Case and -theory suggest that there too it is from the lower head that the -role is assigned.

15

Presumably in much the same way as possessors cross-linguistically are typically not assigned the Case of the

noun they modify.

153

For if the visibility of a -role is dependent on the element carrying that -role receiving Case,
then it is clear that the emphatic pronoun, argued in the last section to be Caseless, cannot be
the bearer of a -role as would be the case if assignment proceeded from Px. If -role
assignment proceeds from v or V, on the other hand, then it will be assigned to Px, and will be
visible by virtue of the fact that this head has Case.

This type of structure, in which a Case- and -marked affix is doubled by a Caseless, non-marked pronoun constitutes a third kind of agreement relation of a kind not attested in Finnish
and two questions now arise, the first pertaining to the precise nature of the relationship
between the emphatic pronoun and the pronominal affix and the second being why pronouns
but not full NP arguments may bear that relationship. By way of answer to the first question,
it is clear that this analysis has much in common with the Pronominal Agreement Hypothesis
as first articulated in generative terms by Jelinek (1984) and elaborated upon in the literature
reviewed in section 3.1 of chapter two, most notably Baker (2003), in which he demonstrates
that pronominal agreement is a property of constructions rather than of languages, and that
where it is present, an overt NP may not occupy the argument position related to that affix16,
but can at most bind the pronominal affix as an adjunct (c.f. Baker, 2003:2). SpecPxP, being a
non-Case- and non--position, has much in common with such an adjoined position, the most
striking difference being that it is located in the middle of the structure, rather than in the
peripheral, dislocated position that such elements occupy in the examples discussed by Baker.
Viewed in this way, it appears that the premise of the second question, that only pronouns
may bear this relationship to an affix, is in fact wrong, since in the multiple subject
construction, discussed in section 4.3.2, broad subjects do seem to bear just this kind of
relation to a Case- and -marked (object) affix and furthermore do occupy a clause-peripheral
position more usually associated with dislocated elements. Since object agreement affixes
only ever appear in positions where they receive a -role, the relationship of the broad subject
hind to the pronominal affix ha in a multiple subject sentence such as (49), repeated here as
(72), must also be one of binding from a non--position.

16

Note that this is also the prediction made by the model being developed here, since it is not possible for both

an agreement affix and an NP argument to receive a -role in a given position. Even in the case of emphatic
pronouns co-occurring with a pronominal affix, independent consideration of the data led to the conclusion that
the doubling element is not located in the specifier of the -assigning head.

154

(72)

hind-uni

yuqbilu-hai

T-Tullb-u

Hind-NOMi meet.3SG.M-3SG.F.ACCi the-students(M)-NOM


I believed Hind to have been met by the students

Why then can such NPs not appear in the same SpecPxP position as emphatic pronominal
arguments? One possible answer to this question can be given in terms of Case, for while
SpecPxP has been argued to be a non-Case position, the (higher) SpecTP position occupied by
broad subjects is a Case position. If the option of remaining Caseless (or of having nominative
by default) were restricted to pronouns, this would explain in a principled way the distribution
of both categories able to double pronominal affixes.

6 OUTSTANDING ISSUES
The preceding discussion of Modern Standard Arabic has shown that, in addition to the two
patterns of agreement attested in Finnish non-finite constructions, it allows two others not
attested in that language. This final section will consider why these configurations are not also
possible in Finnish.
6.1 Why there can be no Expletives in the Finnish Participial Construction
If the structure EXPLVSO, argued to underlie Arabic VSO clauses, were also allowed in
Finnish participial clauses, then the following should be possible.

(73)

*Min halua-n

(sen)

osta-va-nsa sinu-n/Jussi-n/etc.

auto-n

I.NOM want-1SG (EXPL.GEN) buy-VA-3.PX you.SG-GEN/Jussi-GEN car-ACC


I want you/Jussi to buy a car

Since Finnish makes free use of expletives (Holmberg and Nikanne, 2002), it is unlikely to be
the case that the element sen (or its counterpart in any other morphological case) is not
available. However, recall from section 1 of chapter three that sen differs from the other
pronouns in not being able to check the -features of a possessor agreement morpheme and in
this respect behaves like full NP arguments. Even so, there is no reason why a variant of the
same sentence with default agreement should not be possible. (74) shows that it is not.

(74)

*Min halua-n

(sen)

osta-va-n

sinu-n/Jussi-n etc.

auto-n

I.NOM want-1SG (EXPL.GEN) buy-VA-DFT you.SG-GEN/Jussi-GEN car-ACC


I want you/Jussi to buy a car
155

One possible explanation for this comes from the status of the default ending n, which, being
homophonous with the accusative singular ending for NP arguments, is often assumed to
result from Case-marking by the matrix verb (e.g. Koskinen, 1998:3.2.1.1). If this were the
case, then there would be no means of the expletive receiving Case, with the consequence that
it could not be phonetically realised, for the same reasons as it must remain unexpressed in
Arabic clauses not introduced by a Case-assigner. The difference between Arabic and Finnish
in this respect is that while the ultimately null expletive is motivated in Arabic to the extent
that it serves to eliminate the uninterpretable -features of the Px head, there being no such
features to check in Finnish, it is entirely superfluous and hence, by minimalist assumptions,
impossible.

6.2 Why Finnish does not allow Emphatic Doubling of Argumental Affixes
The fact that the category in SpecPxP cannot receive Case in the Finnish participial
construction should not, however, be a barrier to a Caseless emphatic pronoun appearing in
that position, as was argued to be the case for Arabic in section 5. That is to say, there is no
reason why the following should not be possible.

(75)

*Min halua-n

sinu-n17 osta-va-si

auto-n

I.NOM want-1SG you-GEN buy-VA-2SG.PX car-ACC


I want you to buy a car (not Jussi)

Note that, since in this case the participle does not carry the default ending n, it could be
argued (if the proposal mentioned above is correct) that it is not Case-marked by the matrix
verb18. This being the case, there is nothing to prevent haluan assigning Case to the pronoun
(which would then have the form sinut). As such it would then stand in the same relation to
the agreement affix as an exceptionally Case-marked subject in an Arabic SVO clause and
would therefore cause the -features of that affix to delete, under the assumption that the
computational system is blind to the LF-status of the elements upon which it operates (cf.

17

The genitive has been chosen here, since this is the case argued to be the default case for specifiers in Finnish

by Vainikka (1989). The choice of case is academic, however, since pronouns carrying other case endings are
equally ungrammatical.
18

In fact, a more probable explanation is that the Case-marker n is deleted when the possessive suffix is

attached, since this also happens with object nominals, which clearly are in a position to receive Case. The
alternative line of argumentation is pursued here simply in the interests of completeness.

156

sections 1.2 and 3.2.2 of chapter two). Since these -features carry the subject -role (by
virtue of which they are interpretable), deleting them will result in the loss of that semantic
information, ultimately resulting in an LF-crash.

CONCLUSION

The aim of this chapter was to determine whether the mechanisms claimed to underlie the
system of agreement in Finnish non-finite clauses would prove to have more general validity
by applying the model to a historically and areally unrelated language. In some cases, namely
complementiser and object agreement and in SVO clauses, the models could be applied with a
minimum of adjustment. Other clause types, namely those with VSO word-orders
(particularly where the subject is pronominal) did not fit in so well, but by exploring these
cases in detail, it was shown that they could be accommodated and that there were good
reasons why the same structures are not available in Finnish.

157